A Fully Supervised Classification Method of Remote Sensing Images Based on SIFT Training Sample Extraction

Abstract

The invention relates to a remote sensing image full-supervised classification method based on SIFT (Scale Invariant Feature Transform) training sample extraction, comprising the following steps: normalizing the amplitude data of the initial image to obtain a normalized image; using SIFT and Lap The Las detail enhancement operator extracts the key points in the normalized image as candidate samples; calculates the simple texture features of the normalized image, performs threshold segmentation on the gray value of the candidate samples to obtain the training sample coordinates, and the corresponding simple texture The feature is used as a training sample, and the simple texture features of all other normalized images except the training sample are used as samples to be classified, that is, the test sample; the training sample features extracted by SIFT are used to perform SVM (Support Vector Machine) classifier training to generate the optimal classification surface; through the optimal classification surface, the simple texture features of the unknown test samples are classified to obtain the final classification result; the present invention is suitable for surface classification in complex terrain areas and has high classification accuracy. Has a good promotion.

Description

technical field [0001] The invention belongs to the field of remote sensing image processing and pattern recognition, and relates to a remote sensing image full-supervised classification method based on SIFT (Scale Invariant Feature Transform) training sample extraction. Background technique [0002] The essence of remote sensing classification is that the changes in the spectral response of pixels in different regions caused by surface features changing with time or space, to determine whether the ground objects belong to different surface features; to detect the spatial position of different types of surface features; to identify the types and spatial distribution patterns of different surfaces . The classification and analysis technology based on remote sensing images has a wide range of applications, such as urban management planning, land degradation and desertification detection, ocean and inland water body monitoring, and natural disaster prevention and assessment. ...

AI Technical Summary

Problems solved by technology

However, the disadvantage of classifiers such as maximum correlation classifiers is that when the number of samples is high, the complexity of the algorithm is also high, and the optimal classification plane selected by the nearest neighbor classifier is not globally optimal; the classification effect of these classifiers depends on the training samples. The degree of accuracy is high, and it is very difficult to obtain effective labeled samples. The training samples obtained only by threshold segmentation have a low ability to represent test samples.

Images